Skip to main content

Foundational Trials

UK-ROX

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Conservative Oxygen Therapy With Mechanical Ventilation in Critically Ill Adult Patients: the UK-ROX Randomized Clinical Trial
  • Acronym: UK-ROX
  • Year: 2025
  • Journal published in: JAMA
  • Citation: Martin DS, Gould DW, Shahid T, et al; UK-ROX Investigators. Conservative Oxygen Therapy With Mechanical Ventilation in Critically Ill Adult Patients: the UK-ROX Randomized Clinical Trial. JAMA. 2025;334(5):398-408.

Context & Rationale

  • Background
    Oxygen is among the most frequently administered ICU therapies, historically delivered liberally to avoid cellular hypoxia and organ dysfunction.
    Pathophysiological and observational data support biological plausibility for oxygen toxicity (oxidative stress, vasoconstriction, absorption atelectasis) and associations between hyperoxaemia and worse outcomes in mechanically ventilated ICU cohorts.
    Prior adult ICU RCTs comparing conservative vs liberal (or higher) oxygenation targets were generally neutral for mortality, raising the possibility of (i) true absence of benefit, (ii) inadequate between-group oxygen separation, (iii) effect heterogeneity across illness phenotypes and baseline risk, or (iv) outcome selection insensitivity to modest changes in oxygen exposure.
  • Research Question/Hypothesis
    Whether reducing exposure to supplemental oxygen using a protocolised conservative strategy (target SpO2 88%–92%) would reduce 90‑day mortality compared with usual oxygen therapy in invasively ventilated critically ill adults in UK ICUs.
  • Why This Matters
    Even a small absolute survival effect (benefit or harm) would translate to large population impact given the ubiquity of oxygen therapy.
    UK-ROX tests a pragmatic, scalable “treat oxygen like a drug” strategy, embedded in national audit infrastructure, addressing feasibility of very large ICU trials for common supportive-care exposures.

Design & Methods

  • Research Question: In adults receiving unplanned invasive mechanical ventilation and supplemental oxygen in ICU, does conservative oxygen therapy (SpO2 target 90%, range 88%–92%) reduce 90‑day mortality compared with usual oxygen therapy?
  • Study Type: Multicentre, pragmatic, registry-embedded, parallel-group randomised clinical trial across 97 UK ICUs; open-label delivery; centralised outcomes via data linkage; investigator-initiated and publicly funded. 1
  • Population:
    • Setting: adult general ICUs in England, Wales, and Northern Ireland (via ICNARC Case Mix Programme infrastructure).
    • Inclusion: critically ill adults receiving unplanned invasive mechanical ventilation and supplemental oxygen; randomised within 12 hours of meeting eligibility criteria.
    • Key exclusions: clinician judged one strategy indicated/contraindicated; previous UK-ROX enrolment within 90 days; receiving extracorporeal membrane oxygenation; anticipated imminent extubation/discharge/transfer/death/treatment withdrawal (commonly cited under “clinical decision” among non-randomised eligible patients).
  • Intervention:
    • Conservative oxygen therapy: oxygen titrated to achieve SpO2 90% (acceptable range 88%–92%).
    • Mandatory bedside monitor alarms: lower SpO2 alarm 88%; upper alarm 92% (upper alarm could be deactivated if FiO2 = 0.21).
    • Delivered from randomisation until ICU discharge, death, or day 90 (whichever occurred first); if readmitted within 90 days, protocol reinstated.
  • Comparison:
    • Usual oxygen therapy: oxygen administered according to clinician judgement and local practice; no trial-mandated SpO2 target or alarm limits.
    • Co-interventions (ventilation mode, PEEP strategy, sedation, haemodynamic management) were not protocolised and followed local practice.
  • Blinding: Unblinded (non-pharmacologic, bedside titration intervention); primary endpoint (registry-linked mortality) objective; potential for performance bias in co-interventions and event ascertainment for less objective secondary outcomes.
  • Statistics: A total sample size of 16,500 was planned to detect a 2.5% absolute reduction in 90‑day mortality (from 37.0% to 34.5%) with 90% power at a two-sided 5% significance level, allowing for ~6% losses; 2 interim analyses were planned after 4,500 and 10,000 participants using a Haybittle–Peto stopping rule (P<.001) for benefit or harm; primary analysis was intention-to-treat using adjusted models for effect estimation.
  • Follow-Up Period: Primary endpoint at 90 days; additional in-hospital timepoints (ICU and hospital discharge mortality); secondary outcomes included 30-day organ support composite, 60-day mortality, and (for those consenting to linkage) 12‑month mortality and health resource outcomes; supplemental methods detail adherence and oxygen exposure derivations. 2

Key Results

This trial was not stopped early. Two interim analyses (after 4,500 and 10,000 participants) were prespecified with a Haybittle–Peto boundary (P<.001) and the trial proceeded to full recruitment and follow-up completion.

Outcome Conservative oxygen therapy Usual oxygen therapy Effect p value / 95% CI Notes
Primary: Death by 90 days 2908/8211 (35.4%) 2858/8183 (34.9%) Adjusted risk difference 0.7% 95% CI −0.7 to 2.0; P=.28 Primary analysis used linked 90‑day mortality; missingness was low (data unavailable for 40 participants; 66 requested all data removed).
Death at ICU discharge 2388/8258 (28.9%) 2388/8242 (29.0%) Adjusted risk difference −0.1% 95% CI −1.3 to 1.1; P=.94 No signal of early in-ICU mortality benefit or harm.
Death at hospital discharge 2846/8258 (34.5%) 2805/8242 (34.0%) Adjusted risk difference 0.5% 95% CI −0.8 to 1.9; P=.46 Consistent with the 90‑day mortality finding.
Days alive and free of organ support at 30 days (ordinal composite) Mean 12.9 (SD 12.2) Mean 12.8 (SD 12.1) Proportional OR 1.01 95% CI 0.96 to 1.07; P=.64 Mean difference 0.1 days; 95% CI −0.1 to 0.4; P=.30 (reported).
Time to death (to 12 months) — — HR 1.01 95% CI 0.96 to 1.05; P=.82 Secondary analysis in those with longer-term follow-up available.
Oxygen exposure (enhanced data subset): cumulative 100% oxygen-equivalent hours over first 10 days 20.3 h 28.7 h Difference −8.4 h 95% CI −10.8 to −6.0 Estimated 29% lower supplemental oxygen exposure in the conservative group (reported).
Achieved oxygenation (enhanced data subset): mean SpO2 while receiving supplemental oxygen 93.3% 95.2% Difference −2.0% 95% CI −2.2 to −1.8 Mean FiO2 0.31 vs 0.35 (difference −0.04; 95% CI −0.04 to −0.03); mean PaO2 71.5 vs 79.5 mm Hg (difference −8.0; 95% CI −9.3 to −6.7).
Serious adverse events (any) 58/8230 (0.7%) 29/8204 (0.4%) Not reported Not reported Specified SAEs included supraventricular tachycardia 12 vs 8; atrial fibrillation 14 vs 13; myocardial ischaemia/infarction 19 vs 11; mesenteric ischaemia 7 vs 0 (events/patients as reported). 2
  • Despite large sample size and early randomisation (median 5 hours after initiation of invasive ventilation), conservative oxygen therapy did not reduce 90‑day mortality: 35.4% vs 34.9% (adjusted risk difference 0.7%; 95% CI −0.7 to 2.0; P=.28).
  • Protocolisation meaningfully reduced oxygen exposure (29% lower supplemental oxygen exposure; mean SpO2 while on oxygen 93.3% vs 95.2%), but the absolute separation in achieved oxygenation was modest (~2% SpO2 and ~8 mm Hg PaO2 in the enhanced subset).
  • Serious adverse events were uncommon but numerically higher in the conservative group (0.7% vs 0.4%), with rare ischaemic events (e.g., mesenteric ischaemia 7 vs 0) reported in supplemental safety tables.

Internal Validity

  • Randomisation and allocation: Central randomisation with stratification by site and prespecified diagnostic subgroup; allocation concealed until assignment; all 16,500 randomised participants received their allocated strategy as recorded.
  • Post-randomisation exclusions/withdrawal: Requested complete data removal: 28/8258 (0.34%) in conservative vs 38/8242 (0.46%) in usual; primary outcome linkage unavailable for 40 participants (e.g., not an NHS patient); primary analysis included 16,434 participants with analysable 90‑day outcomes.
  • Performance/detection bias: Open-label delivery could influence oxygen titration intensity, alarm responsiveness, and co-interventions; however, the primary endpoint (mortality) is objective and registry-linked, limiting detection bias for the primary outcome.
  • Protocol adherence: In the enhanced subset, non-adherence in the conservative group was frequent by prespecified flags (526/1252 [42.1%] with any non-adherence), though the duration of non-adherence was limited (10.6% of time receiving supplemental oxygen); adherence definitions and monitoring processes were prespecified. 2
  • Baseline characteristics and severity: Randomised groups were well balanced (median age 60 years; female 38.2%); baseline median SpO2 before randomisation was high (97% vs 96%), implying many patients started from a relatively “non-hypoxaemic” baseline oxygenation state.
  • Timing: Median time from initiation of invasive mechanical ventilation in ICU to randomisation was 5 (IQR 2–8) hours, appropriate for an oxygen exposure intervention where early cumulative dose may matter.
  • Dose/separation of variable of interest: Achieved between-group separation (enhanced subset) was consistent but modest: mean FiO2 0.31 vs 0.35 (difference −0.04); mean SpO2 with oxygen 93.3% vs 95.2% (difference −2.0%); mean PaO2 71.5 vs 79.5 mm Hg (difference −8.0); cumulative oxygen exposure over 10 days 20.3 vs 28.7 oxygen-equivalent hours (difference −8.4).
  • Heterogeneity and subgroup credibility: Prespecified subgroup analyses (hypoxic-ischaemic encephalopathy, sepsis, acute brain injury, other) did not show convincing evidence of treatment effect modification (e.g., HIE: 58.0% vs 58.6%, adjusted risk difference −1.3% [−6.2 to 3.6], OR 0.94 [0.76 to 1.17]).
  • Outcome assessment: Mortality was captured via national data linkage (high completeness) and is resistant to ascertainment bias; secondary outcomes relying on linkage/consent were available in subsets and more vulnerable to selection mechanisms.
  • Statistical rigour: Target sample size achieved; interim monitoring prespecified; primary analysis consistent with intention-to-treat and adjusted modelling; no multiplicity adjustment for multiple secondary outcomes (interpretation appropriately cautious).

Conclusion on Internal Validity: Overall, internal validity for the primary endpoint appears strong given concealed randomisation, minimal withdrawal/unlinked outcomes, and objective registry-linked mortality ascertainment; however, open-label care, protocol non-adherence, and modest oxygen separation likely limited the trial’s ability to detect small treatment effects attributable purely to oxygen dose.

External Validity

  • Population representativeness: Large, pragmatic enrolment across 97 UK ICUs with broad unplanned invasive mechanical ventilation eligibility; randomised cohorts were representative of the wider eligible ICU population on key demographic factors (e.g., median age 60 years; female ~38%).
  • Selection considerations: Of 49,233 eligible patients, 16,500 were randomised; 14,865 eligible patients were missed/identified too late; 6,589 did not undergo randomisation due to clinical decision (often imminent extubation/discharge/transfer/death/withdrawal); and 10,582 were excluded because one strategy was indicated or contraindicated by the treating clinician.
  • Applicability across settings: Findings are most directly applicable to ICUs with contemporary oxygen practices similar to the UK (baseline SpO2 before randomisation ~96%–97%); settings with systematically higher baseline oxygenation (or different alarm practices, nurse:patient ratios, and arterial blood gas monitoring) may achieve different separation and potentially different effects.
  • Subpopulation considerations: Patients with clear clinical indications for distinct oxygenation targets (e.g., specific neurological or cardiopulmonary pathophysiology where clinicians deem one strategy necessary) were preferentially excluded by design, limiting inference in those high-stakes phenotypes.

Conclusion on External Validity: Generalisability is high for unplanned invasively ventilated adults in UK-type ICUs and health systems with similar “usual care” oxygen practices, but may be moderate in jurisdictions where usual care is more liberal (or where protocolised conservative titration yields larger oxygen separation).

Strengths & Limitations

  • Strengths: Very large sample size (n=16,500) enabling detection of modest absolute effects; pragmatic registry-embedded delivery enhancing scalability and representativeness; early randomisation (median 5 hours) targeting early oxygen exposure; objective primary outcome via national linkage; prespecified interim monitoring and diagnostic strata.
  • Limitations: Open-label intervention with potential performance bias and unmeasured co-intervention imbalance; modest achieved oxygen separation (particularly given high baseline SpO2), which can dilute treatment effects; notable eligible-but-not-randomised proportion (missed/late identification); secondary outcomes requiring consent for linkage (reduced denominators at 12 months); protocol non-adherence common by trigger criteria; rare but numerically higher serious adverse events in the conservative group warrant cautious interpretation.

Interpretation & Why It Matters

  • Clinical practice implication
    For invasively ventilated adults receiving supplemental oxygen in UK ICUs, targeting SpO2 88%–92% (via alarms and protocolised titration) should not be expected to reduce 90‑day mortality compared with contemporary usual care.
    The pragmatic message is not “oxygen targets do not matter”, but rather that pushing saturations down into a conservative range—from an already relatively moderate usual practice—does not deliver an average mortality benefit at population level.
  • Dose–response framing
    UK-ROX achieved a consistent reduction in oxygen exposure (e.g., FiO2 0.31 vs 0.35; PaO2 71.5 vs 79.5 mm Hg), yet mortality remained unchanged, suggesting either (i) the relevant “harmful” hyperoxia threshold lies higher than typical UK usual care, (ii) benefits are confined to specific subgroups not enriched here, or (iii) modest separation is insufficient to shift hard outcomes.
  • Safety signal interpretation
    Serious adverse events were rare but numerically higher under conservative targets; this does not establish causality, but reinforces the need for vigilance against unintended hypoxaemia (even brief) and careful consideration of patient-specific oxygen delivery constraints.

Controversies & Subsequent Evidence

  • “Usual care” as a moving comparator and oxygen-separation dilution: The accompanying editorial emphasised UK-ROX as an exemplar of large simple trials for ubiquitous supportive-care therapies, while also highlighting the methodological challenge that “usual care” in non-pharmacologic trials can drift and reduce between-group contrast, diluting effects. 34
  • Positioning within the adult ICU RCT landscape: UK-ROX aligns with several adult ICU trials that did not show mortality benefit from conservative oxygenation targets, including ICU-ROX, HOT-ICU, PILOT, ICONIC, and the low-normal vs high-normal oxygenation target trial (organ dysfunction outcome). 56789
  • Context-dependent benefits in specific phenotypes: In COVID-19 ICU patients (HOT-COVID), lower vs higher oxygenation targets increased days alive without life support; and in paediatric ICU (Oxy-PICU), conservative targets improved a 30‑day organ support/death composite—highlighting that effects may differ by disease biology, baseline oxygenation, and outcome choice. 1011
  • Interpretation of earlier “positive” oxygen trials: The Oxygen-ICU trial reported mortality benefit with conservative oxygen therapy but differed in size, setting, and design, and has been considered hypothesis-generating rather than definitive; UK-ROX’s scale and pragmatic delivery substantially upgrades precision around average adult ICU mortality effects in contemporary practice. 12
  • UK baseline oxygen practice and separation feasibility: Pre-UK-ROX observational work in UK ICUs questioned whether oxygenation practice was already relatively “usual” (i.e., not markedly hyperoxaemic), consistent with high baseline SpO2 at enrolment and potentially limiting incremental benefit from further reduction. 13
  • Evidence synthesis post–neutral trials: Contemporary meta-analyses of ICU RCTs have generally reported no significant mortality difference between conservative and liberal oxygen strategies, supporting UK-ROX’s neutral mortality finding while emphasising ongoing uncertainty about subgroup effects and optimal comparator selection. 1415
  • “Heterogeneity of treatment effect” and individualisation: A focused critique argues that averaging across broad ICU populations may obscure meaningful subgroup-specific benefit or harm and advocates for approaches that identify patients most likely to benefit from particular oxygen targets. 16
  • Trial-to-practice translation and next steps: The Mega-ROX programme is designed to address residual uncertainty at scale, including comparator strategies that may ensure clearer oxygen separation (e.g., protocolised liberal oxygen therapy with a minimum acceptable FiO2 of 0.3). 17
  • Secondary analyses suggesting differential effects: Individualised treatment effect modelling across ICU oxygen-target trials suggests patient-level factors may modify outcome effects of oxygen targets, reinforcing the need for stratified/precision approaches rather than a single universal SpO2 target. 18

Summary

  • UK-ROX randomised 16,500 unplanned invasively ventilated adults across 97 UK ICUs to conservative (SpO2 88%–92%) vs usual oxygen therapy.
  • There was no reduction in 90‑day mortality: 35.4% vs 34.9% (adjusted risk difference 0.7%; 95% CI −0.7 to 2.0; P=.28).
  • Conservative therapy reduced oxygen exposure (29% lower oxygen-equivalent hours; mean SpO2 on oxygen 93.3% vs 95.2%), but achieved separation was modest.
  • Secondary outcomes (organ support–free days; ICU/hospital discharge mortality; time to death) were similarly neutral.
  • Serious adverse events were rare but numerically higher under conservative targets (0.7% vs 0.4%); uncommon ischaemic events were reported in supplemental tables and should be monitored in future programmes.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • For historical context on conservative oxygen in ICU and ongoing debate about early “positive” signals, see Oxygen-ICU (JAMA 2016) and subsequent larger pragmatic ICU trials.
  • UK-ROX’s registry-embedded approach can be read alongside broader methodological calls for integration of clinical trials with routine care systems.

Overall Takeaway

UK-ROX is a landmark pragmatic ICU trial because it demonstrates—at national scale and with high-fidelity registry linkage—that protocolising a conservative SpO2 target (88%–92%) reduces oxygen exposure but does not improve 90‑day mortality compared with contemporary UK usual care. The trial shifts the field from “should we be conservative?” to the harder questions of which patients, how much separation, and which comparators and outcomes are needed to detect clinically meaningful oxygen-treatment effects.

Overall Summary

  • Conservative oxygen titration (SpO2 88%–92%) lowered oxygen exposure but did not reduce 90‑day mortality vs usual care in 16,500 ventilated UK ICU adults.
  • Achieved oxygen separation was real but modest, a likely contributor to neutral clinical outcomes in a setting where baseline oxygenation was already relatively high.
  • Future work is best directed at comparator optimisation (ensuring separation) and identifying patient-level heterogeneity rather than assuming a single universal oxygen target.

Bibliography

  • Martin DS, Shahid T, Gould DW, et al. Evaluating the clinical and cost-effectiveness of a conservative approach to oxygen therapy for invasively ventilated adults in intensive care: protocol for the UK-ROX trial. J Intensive Care Soc. 2024;25(2):223-230. Link
  • Martin DS, Gould DW, Shahid T, et al; UK-ROX Investigators. Supplementary content (eMethods/eTables) for: Conservative oxygen therapy in mechanically ventilated critically ill adult patients: the UK-ROX randomized clinical trial. JAMA. 2025. Link
  • Seitz KP, Hiemstra B, Wiersinga WJ. Patient, Treatment, Outcome—Large Simple Trials of Common Therapies. JAMA. 2025;334(5):395-397. Link
  • Thompson BT, Schoenfeld D. Usual care as the control group in clinical trials of nonpharmacologic interventions. Proc Am Thorac Soc. 2007;4(7):577-582. Link
  • Mackle D, Bellomo R, Bailey M, et al; ICU-ROX Investigators and the Australian and New Zealand Intensive Care Society Clinical Trials Group. Conservative oxygen therapy during mechanical ventilation in the ICU. N Engl J Med. 2020;382(11):989-998. Link
  • Schjørring OL, Klitgaard TL, Perner A, et al; HOT-ICU Investigators. Lower or higher oxygenation targets for acute hypoxemic respiratory failure. N Engl J Med. 2021;384(14):1301-1311. Link
  • Semler MW, Casey JD, Lloyd BD, et al; PILOT Investigators and the Pragmatic Critical Care Research Group. Oxygen-saturation targets for critically ill adults receiving mechanical ventilation. N Engl J Med. 2022;387(19):1759-1769. Link
  • van der Wal LI, Grim CCA, Del Prado MR, et al; ICONIC Investigators. Conservative versus liberal oxygenation targets in intensive care unit patients (ICONIC): a randomized clinical trial. Am J Respir Crit Care Med. 2023;208(7):770-779. Link
  • Gelissen H, de Grooth HJ, Smulders Y, et al. Effect of low-normal vs high-normal oxygenation targets on organ dysfunction in critically ill patients: a randomized clinical trial. JAMA. 2021;326(10):940-948. Link
  • Nielsen FM, Klitgaard TL, Siegemund M, et al; HOT-COVID Trial Group. Lower vs higher oxygenation target and days alive without life support in COVID-19: the HOT-COVID randomized clinical trial. JAMA. 2024;331(14):1185-1194. Link
  • Peters MJ, Gould DW, Ray S, et al; Oxy-PICU Investigators of the Paediatric Critical Care Society Study Group (PCCS-SG). Conservative versus liberal oxygenation targets in critically ill children (Oxy-PICU): a UK multicentre, open, parallel-group, randomised clinical trial. Lancet. 2024;403(10424):355-364. Link
  • Girardis M, Busani S, Damiani E, et al. Effect of conservative vs conventional oxygen therapy on mortality among patients in an intensive care unit: the Oxygen-ICU randomized clinical trial. JAMA. 2016;316(15):1583-1589. Link
  • Post B, Palmer E, Harris S, Singer M, Martin D. Oxygenation of the critically ill in selected intensive care units in the UK: are we usual? Br J Anaesth. 2020;125(3):e277-e279. Link
  • Li XY, Dai B, Hou HJ, et al. Conservative versus liberal oxygen therapy for intensive care unit patients: meta-analysis of randomized controlled trials. Ann Intensive Care. 2024;14(1):68. Link
  • Martin DS, Mckenna HT, Rowan KM, et al. The effect of conservative oxygen therapy on mortality in adult critically ill patients: a systematic review and meta-analysis of randomised controlled trials. J Intensive Care Soc. 2023;24(4):399-408. Link
  • Martin DS, Grocott MPW. Heterogeneity of treatment effect: the case for individualising oxygen therapy in critically ill patients. Crit Care. 2025;29(1):50. Link
  • Young PJ, Arabi YM, Bagshaw SM, et al; Mega-ROX Management Committee; Australian and New Zealand Intensive Care Society Clinical Trials Group; Crit Care Asia and Africa Network; Irish Critical Care Clinical Trials Group; Alberta Health Services Critical Care Strategic Clinical Network. Protocol and statistical analysis plan for the mega randomised registry trial research program comparing conservative versus liberal oxygenation targets in adults receiving unplanned invasive mechanical ventilation in the ICU (Mega-ROX). Crit Care Resusc. 2023;24(2):137-149. Link
  • Buell KG, Spicer AB, Casey JD, et al. Individualized treatment effects of oxygen targets in mechanically ventilated critically ill adults. JAMA. 2024;331(14):1195-1204. Link